For twenty years a voter suppression campaign has crept through the US, keeping millions from the polls. Unlike obvious forms of disenfranchisement, it doesn’t discriminate by race, class, or party– but by voting history.
Starting with Bush 2000 and exploding with social media, campaigns’ winning strategy has been to target and turn out only reliable voters who already support their candidate. People with consistent voting histories get Facebook ads, postcards, and canvassers at their door pleading for their vote. First-time or sporadic voters don’t.
It’s a vicious cycle: low turnout voters don’t hear from campaigns; they vote less; they don’t get targeted by future campaigns.
In theory, candidates who reach voters beyond their hyper-engaged base gain an edge over their opponents and break the cycle. Why doesn’t this happen?
It’s easy to identify your base; it’s much harder to identify sympathetic people who don’t vote, but might be persuaded to. Voter targeting has become commodified by a handful of big data providers2Almost no campaign has the resources, time, or expertise to attempt their own modeling. Even fewer campaign managers would risk their necks by experimenting. No one gets fired for choosing IBM; no one gets fired for using Civis scores. offer campaigns two key metrics on each voter: “likelihood to vote” and “likelihood of support”.
These metrics are typically identical across all candidates of the same party. A campaign can target voters who lean Democrat and vote often, but when it comes to organizing the sort of person-to-person outreach most likely to sway voter behavior, this partisanship + turnout model is literally two-dimensional.
For example, a 22-year-old student who didn’t vote in 2016 would be labeled “low-turnout” by these models and ignored. But if that same student has several high-turnout friends, reaching out through those politically engaged connections has a good chance of influencing them to vote. Persuadability depends on who’s doing the persuading; peer pressure succeeds where civic duty falls short.
While traditional microtargeting uses demographic data to tailor its message to individuals, “relational” targeting must also choose its messenger. The sort of social relationship data for such an undertaking is mostly owned by Facebook and Google, who guard it jealously.
But campaigns have voter data, and volunteers know voters. People have, on average, 500-2000 phone and email contacts. A campaign’s 20 volunteers might be connected to 20,000 voters– impressive, but too many to contact. Stratos combines voter modeling with social network analysis to prioritize the most persuadable voters and pair them with the volunteers who know them best.
Microtargeting, custom ads, and personalized new feeds isolate us and make us vulnerable to disinformation and division.4For a deep dive into the broken promises and growing perils of microtargeting, see https://www.ivir.nl/publicaties/download/UtrechtLawReview.pdf Relational targeting can break us out of our new feeds and get us talking to each other again. What’s more, it changes campaigns’ competitive calculus. To win, campaigns will need to care about voters beyond their base again.
Notes [ + ]
|1.||↑|| “Voter Turnout Demographics – United States Elections Project.” Accessed July 4, 2020. http://www.electproject.org/home/voter-turnout/demographics.|
|2.||↑||Almost no campaign has the resources, time, or expertise to attempt their own modeling. Even fewer campaign managers would risk their necks by experimenting. No one gets fired for choosing IBM; no one gets fired for using Civis scores.|
|3.||↑|| Ruffini, Patrick. “The Myth of the White Suburban Swing Voter.” Medium, June 15, 2020. https://medium.com/@PatrickRuffini/the-myth-of-the-white-suburban-swing-voter-581f1f23faba.|
|4.||↑||For a deep dive into the broken promises and growing perils of microtargeting, see https://www.ivir.nl/publicaties/download/UtrechtLawReview.pdf|